On Mon, 2022-03-28 at 09:34:35 +0000, Éric Valette wrote: > On 3/28/22 01:29, Guillem Jover wrote: > >> Invoquing the dkms postinstall script fails because it does not find > >> /lib/modules/x.y.z > >> > >> directories and rebooting without graphic drivers and other stff does not > >> work.
> > Just to understand, did dpkg-fsys-usrunmess fail (exited non-zero) while > > > > configuring packages (with dkms)? And then you rebooted? > There was no printed error beside the arror from dkms postinst (virtualbox > assumed it was running virtualized, and nvidia just failling): just a > congratulation message. Diid not check the return value of the script via > echo $? Ok, so if the script did not finish with something like: ,--- error: cannot reconfigure packages: ... `--- Then I suppose it exited normally with something like: ,--- Done, hierarchy unmessed, congrats! Rebooting now is very strongly advised. (Note: you might need to run 'hash -r' in your shell.) `--- ? If the latter, I guess the maintainer script(s) running dkms, didn't exit non-zero. Hmm. Where those from external third-party packages? > >> So you probably should copy /usr/lib/modules to /lib/modules before the > >> calling the > >> postinstall. > > Hmm, right, because these were not tracked, they got missed in the > > migration. I've checked the Debian archive and at least there, it does > > not look like anything else besides kernel modules are being current > > shipped in those directories (via apt-file), but the problem is that > > I've seen references in source code (via codesearch.d.o) to > > /usr/lib/modules, for at least apache and python modules, so I don't > > think an unconditional move for untracked files would be safe there. > > I'm thinking the following special-case options (in order of decreasing > > preference): > > > > * move only untracked «/usr/lib/modules/[0-9]*» expecting/assuming > > those to be kernel modules, > > You could move only the current running kernel modules using uname > to find the version. It will not hurt and they will be rebuild anyway at > least for the dkms ones. I guess that would be another option yes, thanks. > The real problem being that dkms generated file does not appear in any file Sure. I'll try to prepare something later today. Thanks, Guillem